Prediction of Anti-Glioblastoma Drug-Decorated Nanoparticle Delivery Systems Using Molecular Descriptors and Machine Learning.

The theoretical prediction of drug-decorated nanoparticles (DDNPs) has become a very important task in medical applications. For the current paper, Perturbation Theory Machine Learning (PTML) models were built to predict the probability of different pairs of drugs and nanoparticles creating DDNP complexes with anti-glioblastoma activity. PTML models use the perturbations of molecular descriptors of drugs and nanoparticles as inputs in experimental conditions. The raw dataset was obtained by mixing the nanoparticle experimental data with drug assays from the ChEMBL database. Ten types of machine learning methods have been tested. Only 41 features have been selected for 855,129 drug-nanoparticle complexes. The best model was obtained with the Bagging classifier, an ensemble meta-estimator based on 20 decision trees, with an area under the receiver operating characteristic curve (AUROC) of 0.96, and an accuracy of 87% (test subset). This model could be useful for the virtual screening of nanoparticle-drug complexes in glioblastoma. All the calculations can be reproduced with the datasets and python scripts, which are freely available as a GitHub repository from authors.

Prediction of Antimalarial Drug-Decorated Nanoparticle Delivery Systems with Random Forest Models.

Drug-decorated nanoparticles (DDNPs) have important medical applications. The current work combined Perturbation Theory with Machine Learning and Information Fusion (PTMLIF). Thus, PTMLIF models were proposed to predict the probability of nanoparticle-compound/drug complexes having antimalarial activity (against Plasmodium). The aim is to save experimental resources and time by using a virtual screening for DDNPs. The raw data was obtained by the fusion of experimental data for nanoparticles with compound chemical assays from the ChEMBL database. The inputs for the eight Machine Learning classifiers were transformed features of drugs/compounds and nanoparticles as perturbations of molecular descriptors in specific experimental conditions (experiment-centered features). The resulting dataset contains 107 input features and 249,992 examples. The best classification model was provided by Random Forest, with 27 selected features of drugs/compounds and nanoparticles in all experimental conditions considered. The high performance of the model was demonstrated by the mean Area Under the Receiver Operating Characteristics (AUC) in a test subset with a value of 0.9921 ± 0.000244 (10-fold cross-validation). The results demonstrated the power of information fusion of the experimental-centered features of drugs/compounds and nanoparticles for the prediction of nanoparticle-compound antimalarial activity. The scripts and dataset for this project are available in the open GitHub repository.

Developing a Secure Low-Cost Radon Monitoring System.

Radon gas has been declared a human carcinogen by the United States Environmental Protection Agency (USEPA) and the International Agency for Research on Cancer (IARC). Several studies carried out in Spain highlighted the high radon concentrations in several regions, with Galicia (northwestern Spain) being one of the regions with the highest radon concentrations. The objective of this work was to create a safe and low-cost radon monitoring and alert system, based on open source technologies. To achieve this objective, the system uses devices, a collection of sensors with a processing unit and a communication module, and a backend, responsible for managing all the information, predicting radon levels and issuing alerts using open source technologies. Security is one of the largest challenges for the internet of things, and it is utterly important in the current scenario, given that high radon concentrations pose a health risk. For this reason, this work focuses on securing the entire end-to-end communication path to avoid data forging. The results of this work indicate that the development of a low-cost, yet secured, radon monitoring system is feasible, allowing one to create a network of sensors that can help mitigate the health hazards that high radon concentrations pose.

Improvement of Epitope Prediction Using Peptide Sequence Descriptors and Machine Learning.

In this work, we improved a previous model used for the prediction of proteomes as new B-cell epitopes in vaccine design. The predicted epitope activity of a queried peptide is based on its sequence, a known reference epitope sequence under specific experimental conditions. The peptide sequences were transformed into molecular descriptors of sequence recurrence networks and were mixed under experimental conditions. The new models were generated using 709,100 instances of pair descriptors for query and reference peptide sequences. Using perturbations of the initial descriptors under sequence or assay conditions, 10 transformed features were used as inputs for seven Machine Learning methods. The best model was obtained with random forest classifiers with an Area Under the Receiver Operating Characteristics (AUROC) of 0.981 ± 0.0005 for the external validation series (five-fold cross-validation). The database included information about 83,683 peptides sequences, 1448 epitope organisms, 323 host organisms, 15 types of in vivo processes, 28 experimental techniques, and 505 adjuvant additives. The current model could improve the in silico predictions of epitopes for vaccine design. The script and results are available as a free repository.

A methodology for the design of experiments in computational intelligence with multiple regression models.

The design of experiments and the validation of the results achieved with them are vital in any research study. This paper focuses on the use of different Machine Learning approaches for regression tasks in the field of Computational Intelligence and especially on a correct comparison between the different results provided for different methods, as those techniques are complex systems that require further study to be fully understood. A methodology commonly accepted in Computational intelligence is implemented in an R package called RRegrs. This package includes ten simple and complex regression models to carry out predictive modeling using Machine Learning and well-known regression algorithms. The framework for experimental design presented herein is evaluated and validated against RRegrs. Our results are different for three out of five state-of-the-art simple datasets and it can be stated that the selection of the best model according to our proposal is statistically significant and relevant. It is of relevance to use a statistical approach to indicate whether the differences are statistically significant using this kind of algorithms. Furthermore, our results with three real complex datasets report different best models than with the previously published methodology. Our final goal is to provide a complete methodology for the use of different steps in order to compare the results obtained in Computational Intelligence problems, as well as from other fields, such as for bioinformatics, cheminformatics, etc., given that our proposal is open and modifiable.

Texture analysis in gel electrophoresis images using an integrative kernel-based approach.

Texture information could be used in proteomics to improve the quality of the image analysis of proteins separated on a gel. In order to evaluate the best technique to identify relevant textures, we use several different kernel-based machine learning techniques to classify proteins in 2-DE images into spot and noise. We evaluate the classification accuracy of each of these techniques with proteins extracted from ten 2-DE images of different types of tissues and different experimental conditions. We found that the best classification model was FSMKL, a data integration method using multiple kernel learning, which achieved AUROC values above 95% while using a reduced number of features. This technique allows us to increment the interpretability of the complex combinations of textures and to weight the importance of each particular feature in the final model. In particular the Inverse Difference Moment exhibited the highest discriminating power. A higher value can be associated with an homogeneous structure as this feature describes the homogeneity; the larger the value, the more symmetric. The final model is performed by the combination of different groups of textural features. Here we demonstrated the feasibility of combining different groups of textures in 2-DE image analysis for spot detection.

Improving enzyme regulatory protein classification by means of SVM-RFE feature selection.

Enzyme regulation proteins are very important due to their involvement in many biological processes that sustain life. The complexity of these proteins, the impossibility of identifying direct quantification molecular properties associated with the regulation of enzymatic activities, and their structural diversity creates the necessity for new theoretical methods that can predict the enzyme regulatory function of new proteins. The current work presents the first classification model that predicts protein enzyme regulators using the Markov mean properties. These protein descriptors encode the topological information of the amino acid into contact networks based on amino acid distances and physicochemical properties. MInD-Prot software calculated these molecular descriptors for 2415 protein chains (350 enzyme regulators) using five atom physicochemical properties (Mulliken electronegativity, Kang-Jhon polarizability, vdW area, atom contribution to P) and the protein 3D regions. The best classification models to predict enzyme regulators have been obtained with machine learning algorithms from Weka using 18 features. K* has been demonstrated to be the most accurate algorithm for this protein function classification. Wrapper Subset Evaluator and SVM-RFE approaches were used to perform a feature subset selection with the best results obtained from SVM-RFE. Classification performance employing all the available features can be reached using only the 8 most relevant features selected by SVM-RFE. Thus, the current work has demonstrated the possibility of predicting new molecular targets involved in enzyme regulation using fast theoretical algorithms.

Markov mean properties for cell death-related protein classification.

The cell death (CD) is a dynamic biological function involved in physiological and pathological processes. Due to the complexity of CD, there is a demand for fast theoretical methods that can help to find new CD molecular targets. The current work presents the first classification model to predict CD-related proteins based on Markov Mean Properties. These protein descriptors have been calculated with the MInD-Prot tool using the topological information of the amino acid contact networks of the 2423 protein chains, five atom physicochemical properties and the protein 3D regions. The Machine Learning algorithms from Weka were used to find the best classification model for CD-related protein chains using all 20 attributes. The most accurate algorithm to solve this problem was K*. After several feature subset methods, the best model found is based on only 11 variables and is characterized by the Area Under the Receiver Operating Characteristic Curve (AUROC) of 0.992 and the true positive rate (TP Rate) of 88.2% (validation set). 7409 protein chains labeled with "unknown function" in the PDB Databank were analyzed with the best model in order to predict the CD-related biological activity. Thus, several proteins have been predicted to have CD-related function in Homo sapiens: 3DRX-involved in virus-host interaction biological process, protein homooligomerization; 4DWF-involved in cell differentiation, chromatin modification, DNA damage response, protein stabilization; 1IUR-involved in ATP binding, chaperone binding; 1J7D-involved in DNA double-strand break processing, histone ubiquitination, nucleotide-binding oligomerization; 1UTU-linked with DNA repair, regulation of transcription; 3EEC-participating to the cellular membrane organization, egress of virus within host cell, class mediator resulting in cell cycle arrest, negative regulation of ubiquitin-protein ligase activity involved in mitotic cell cycle and apoptotic process. Other proteins from bacteria predicted as CD-related are 2G3V - a CAG pathogenicity island protein 13 from Helicobacter pylori, 4G5A - a hypothetical protein in Bacteroides thetaiotaomicron, 1YLK-involved in the nitrogen metabolism of Mycobacterium tuberculosis, and 1XSV - with possible DNA/RNA binding domains. The results demonstrated the possibility to predict CD-related proteins using molecular information encoded into the protein 3D structure. Thus, the current work demonstrated the possibility to predict new molecular targets involved in cell-death processes.

Kernel-based feature selection techniques for transport proteins based on star graph topological indices.

The transport of the molecules inside cells is a very important topic, especially in Drug Metabolism. The experimental testing of the new proteins for the transporter molecular function is expensive and inefficient due to the large amount of new peptides. Therefore, there is a need for cheap and fast theoretical models to predict the transporter proteins. In the current work, the primary structure of a protein is represented as a molecular Star graph, characterized by a series of topological indices. The dataset was made up of 2,503 protein chains, out of which 413 have transporter molecular function and 2,090 have no transporter function. These indices were used as input to several classification techniques to find the best Quantitative Structure Activity Relationship (QSAR) model that can evaluate the transporter function of a new protein chain. Among several feature selection techniques, the Support Vector Machine Recursive Feature Elimination allows us to obtain a classification model based on 20 attributes with a true positive rate of 83% and a false positive rate of 16.7%.

Evolutionary computation and QSAR research.

The successful high throughput screening of molecule libraries for a specific biological property is one of the main improvements in drug discovery. The virtual molecular filtering and screening relies greatly on quantitative structure-activity relationship (QSAR) analysis, a mathematical model that correlates the activity of a molecule with molecular descriptors. QSAR models have the potential to reduce the costly failure of drug candidates in advanced (clinical) stages by filtering combinatorial libraries, eliminating candidates with a predicted toxic effect and poor pharmacokinetic profiles, and reducing the number of experiments. To obtain a predictive and reliable QSAR model, scientists use methods from various fields such as molecular modeling, pattern recognition, machine learning or artificial intelligence. QSAR modeling relies on three main steps: molecular structure codification into molecular descriptors, selection of relevant variables in the context of the analyzed activity, and search of the optimal mathematical model that correlates the molecular descriptors with a specific activity. Since a variety of techniques from statistics and artificial intelligence can aid variable selection and model building steps, this review focuses on the evolutionary computation methods supporting these tasks. Thus, this review explains the basic of the genetic algorithms and genetic programming as evolutionary computation approaches, the selection methods for high-dimensional data in QSAR, the methods to build QSAR models, the current evolutionary feature selection methods and applications in QSAR and the future trend on the joint or multi-task feature selection methods.

Applied computational techniques on schizophrenia using genetic mutations.

Schizophrenia is a complex disease, with both genetic and environmental influence. Machine learning techniques can be used to associate different genetic variations at different genes with a (schizophrenic or non-schizophrenic) phenotype. Several machine learning techniques were applied to schizophrenia data to obtain the results presented in this study. Considering these data, Quantitative Genotype - Disease Relationships (QDGRs) can be used for disease prediction. One of the best machine learning-based models obtained after this exhaustive comparative study was implemented online; this model is an artificial neural network (ANN). Thus, the tool offers the possibility to introduce Single Nucleotide Polymorphism (SNP) sequences in order to classify a patient with schizophrenia. Besides this comparative study, a method for variable selection, based on ANNs and evolutionary computation (EC), is also presented. This method uses half the number of variables as the original ANN and the variables obtained are among those found in other publications. In the future, QDGR models based on nucleic acid information could be expanded to other diseases.

Exploring patterns of epigenetic information with data mining techniques.

Data mining, a part of the Knowledge Discovery in Databases process (KDD), is the process of extracting patterns from large data sets by combining methods from statistics and artificial intelligence with database management. Analyses of epigenetic data have evolved towards genome-wide and high-throughput approaches, thus generating great amounts of data for which data mining is essential. Part of these data may contain patterns of epigenetic information which are mitotically and/or meiotically heritable determining gene expression and cellular differentiation, as well as cellular fate. Epigenetic lesions and genetic mutations are acquired by individuals during their life and accumulate with ageing. Both defects, either together or individually, can result in losing control over cell growth and, thus, causing cancer development. Data mining techniques could be then used to extract the previous patterns. This work reviews some of the most important applications of data mining to epigenetics.

LIBP-Pred: web server for lipid binding proteins using structural network parameters; PDB mining of human cancer biomarkers and drug targets in parasites and bacteria.

Lipid-Binding Proteins (LIBPs) or Fatty Acid-Binding Proteins (FABPs) play an important role in many diseases such as different types of cancer, kidney injury, atherosclerosis, diabetes, intestinal ischemia and parasitic infections. Thus, the computational methods that can predict LIBPs based on 3D structure parameters became a goal of major importance for drug-target discovery, vaccine design and biomarker selection. In addition, the Protein Data Bank (PDB) contains 3000+ protein 3D structures with unknown function. This list, as well as new experimental outcomes in proteomics research, is a very interesting source to discover relevant proteins, including LIBPs. However, to the best of our knowledge, there are no general models to predict new LIBPs based on 3D structures. We developed new Quantitative Structure-Activity Relationship (QSAR) models based on 3D electrostatic parameters of 1801 different proteins, including 801 LIBPs. We calculated these electrostatic parameters with the MARCH-INSIDE software and they correspond to the entire protein or to specific protein regions named core, inner, middle, and surface. We used these parameters as inputs to develop a simple Linear Discriminant Analysis (LDA) classifier to discriminate 3D structure of LIBPs from other proteins. We implemented this predictor in the web server named LIBP-Pred, freely available at , along with other important web servers of the Bio-AIMS portal. The users can carry out an automatic retrieval of protein structures from PDB or upload their custom protein structural models from their disk created with LOMETS server. We demonstrated the PDB mining option performing a predictive study of 2000+ proteins with unknown function. Interesting results regarding the discovery of new Cancer Biomarkers in humans or drug targets in parasites have been discussed here in this sense.